Ryan Joseph

@umdcs '29 | @databallr | @Shot_Quality

[Ryan Joseph @rjanalytics7002]

Here’s a look at Brock Purdy’s passing heatmap through week 3. Purdy has loved the addition of Mike Evans so far, frequently targeting him on the left side of the field.

[🕸️ @biasedninersfan]

This is the MVP of our league, take a bow

[Ryan Joseph @rjanalytics7002]

@TerpsFootball

[databallr @databallr]

dear nba 2k world,

databallr ISO is live

https://t.co/pWCUGBh817

1v1 basketball game to 11. 2s and 3s coming.
leaderboards elo, tournaments, rapm

built with @threejs claude opus 5.5

[Ryan Joseph @rjanalytics7002]

As a preview for tonight's LSU - Ole Miss game, here's how Sam Leavitt and Trinidad Chambliss looked last week against their respective opponents.

[Ryan Joseph @rjanalytics7002]

Here’s a look at how Grunkemeyer and Washington looked last week.

[CFB_Wyatt @wyattwendry2023]

CFB will be woken up on Saturday night. Malik Washington is him. Terps by double digits

[Ryan Joseph @rjanalytics7002]

Here's a closer look at Trevor Lawrence's throws from week 1. Lawrence led the NFL in CPOE in week 1 according to @NextGenStats.

[ShotQuality @Shot_Quality]

Average receiver separation vs completion percentage from week 1

[ShotQuality @Shot_Quality]

Average receiver separation vs completion percentage from week 1

[Ryan Joseph @rjanalytics7002]

@funakistats really cool

[databallr @databallr]

I have another Six Rings challenge for ball-knowers out there!

25 squares. Pick players whose DARKO impact beats their row’s cost. Miss and you get a strike.

Clear the cube before your third strike to win.

https://t.co/sk9PJ8PLkK

[Ryan Joseph @rjanalytics7002]

Had to take down the earlier post. I really appreciate all the support on it and excited to share new work soon!

[Ryan Joseph @rjanalytics7002]

@NigelHiggs7 You use homography to map players. There is a model that identifies reference points on the court in the videos to allow the player locations to be transformed onto a 2D map. They have to be touching the ground though, which is why the ball can’t be reliably mapped all the time.

[Ryan Joseph @rjanalytics7002]

@GoranMandic1717 Roboflow has worked on soccer tracking. I was inspired by their stuff to make this.

https://t.co/CkXX9nWaXK

[Ryan Joseph @rjanalytics7002]

@DannyDimes1000 Data is from @Shot_Quality

[Ryan Joseph @rjanalytics7002]

@GoranMandic1717 Yeah for sure. There is plenty of public soccer tracking data out there to do the same thing.

[Ryan Joseph @rjanalytics7002]

@makingtronics Yeah that’s basically what I’m doing right now

[Ryan Joseph @rjanalytics7002]

@DerekAustin33 I’m looking to retrain the model to detect shots and use that predicted frame

[Ryan Joseph @rjanalytics7002]

Current limitations are that the current model only uses geometric features and does not factor in individual player ability which is something I'm looking to add in the future. It also struggles with plays where players are crowded and when the ball is hidden from view.

[Ryan Joseph @rjanalytics7002]

It uses an expected points model trained on shots that tracked the court positioning of all 10 players. A computer vision model tracks the players and the ball throughout a play. The player locations are mapped onto a 2D court which is then inputted into the expected points model

[Ryan Joseph @rjanalytics7002]

As a follow up, here is eFG% allowed when defending 2026 NBA draft picks.

[Ryan Joseph @rjanalytics7002]

eFG% allowed when defending high major all-conference players last season.